1/* 2 * linux/boot/head.S 3 * 4 * Copyright (C) 1991, 1992, 1993 Linus Torvalds 5 */ 6 7/* 8 * head.S contains the 32-bit startup code. 9 * 10 * NOTE!!! Startup happens at absolute address 0x00001000, which is also where 11 * the page directory will exist. The startup code will be overwritten by 12 * the page directory. [According to comments etc elsewhere on a compressed 13 * kernel it will end up at 0x1000 + 1Mb I hope so as I assume this. - AC] 14 * 15 * Page 0 is deliberately kept safe, since System Management Mode code in 16 * laptops may need to access the BIOS data stored there. This is also 17 * useful for future device drivers that either access the BIOS via VM86 18 * mode. 19 */ 20 21/* 22 * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996 23 */ 24 .code32 25 .text 26 27#include <linux/init.h> 28#include <linux/linkage.h> 29#include <asm/segment.h> 30#include <asm/pgtable_types.h> 31#include <asm/page_types.h> 32#include <asm/boot.h> 33#include <asm/msr.h> 34#include <asm/processor-flags.h> 35#include <asm/asm-offsets.h> 36 37 __HEAD 38 .code32 39ENTRY(startup_32) 40 cld 41 /* 42 * Test KEEP_SEGMENTS flag to see if the bootloader is asking 43 * us to not reload segments 44 */ 45 testb $(1<<6), BP_loadflags(%esi) 46 jnz 1f 47 48 cli 49 movl $(__KERNEL_DS), %eax 50 movl %eax, %ds 51 movl %eax, %es 52 movl %eax, %ss 531: 54 55/* 56 * Calculate the delta between where we were compiled to run 57 * at and where we were actually loaded at. This can only be done 58 * with a short local call on x86. Nothing else will tell us what 59 * address we are running at. The reserved chunk of the real-mode 60 * data at 0x1e4 (defined as a scratch field) are used as the stack 61 * for this calculation. Only 4 bytes are needed. 62 */ 63 leal (BP_scratch+4)(%esi), %esp 64 call 1f 651: popl %ebp 66 subl $1b, %ebp 67 68/* setup a stack and make sure cpu supports long mode. */ 69 movl $boot_stack_end, %eax 70 addl %ebp, %eax 71 movl %eax, %esp 72 73 call verify_cpu 74 testl %eax, %eax 75 jnz no_longmode 76 77/* 78 * Compute the delta between where we were compiled to run at 79 * and where the code will actually run at. 80 * 81 * %ebp contains the address we are loaded at by the boot loader and %ebx 82 * contains the address where we should move the kernel image temporarily 83 * for safe in-place decompression. 84 */ 85 86#ifdef CONFIG_RELOCATABLE 87 movl %ebp, %ebx 88 movl BP_kernel_alignment(%esi), %eax 89 decl %eax 90 addl %eax, %ebx 91 notl %eax 92 andl %eax, %ebx 93#else 94 movl $LOAD_PHYSICAL_ADDR, %ebx 95#endif 96 97 /* Target address to relocate to for decompression */ 98 addl $z_extract_offset, %ebx 99 100/* 101 * Prepare for entering 64 bit mode 102 */ 103 104 /* Load new GDT with the 64bit segments using 32bit descriptor */ 105 leal gdt(%ebp), %eax 106 movl %eax, gdt+2(%ebp) 107 lgdt gdt(%ebp) 108 109 /* Enable PAE mode */ 110 xorl %eax, %eax 111 orl $(X86_CR4_PAE), %eax 112 movl %eax, %cr4 113 114 /* 115 * Build early 4G boot pagetable 116 */ 117 /* Initialize Page tables to 0 */ 118 leal pgtable(%ebx), %edi 119 xorl %eax, %eax 120 movl $((4096*6)/4), %ecx 121 rep stosl 122 123 /* Build Level 4 */ 124 leal pgtable + 0(%ebx), %edi 125 leal 0x1007 (%edi), %eax 126 movl %eax, 0(%edi) 127 128 /* Build Level 3 */ 129 leal pgtable + 0x1000(%ebx), %edi 130 leal 0x1007(%edi), %eax 131 movl $4, %ecx 1321: movl %eax, 0x00(%edi) 133 addl $0x00001000, %eax 134 addl $8, %edi 135 decl %ecx 136 jnz 1b 137 138 /* Build Level 2 */ 139 leal pgtable + 0x2000(%ebx), %edi 140 movl $0x00000183, %eax 141 movl $2048, %ecx 1421: movl %eax, 0(%edi) 143 addl $0x00200000, %eax 144 addl $8, %edi 145 decl %ecx 146 jnz 1b 147 148 /* Enable the boot page tables */ 149 leal pgtable(%ebx), %eax 150 movl %eax, %cr3 151 152 /* Enable Long mode in EFER (Extended Feature Enable Register) */ 153 movl $MSR_EFER, %ecx 154 rdmsr 155 btsl $_EFER_LME, %eax 156 wrmsr 157 158 /* 159 * Setup for the jump to 64bit mode 160 * 161 * When the jump is performend we will be in long mode but 162 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1 163 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use 164 * the new gdt/idt that has __KERNEL_CS with CS.L = 1. 165 * We place all of the values on our mini stack so lret can 166 * used to perform that far jump. 167 */ 168 pushl $__KERNEL_CS 169 leal startup_64(%ebp), %eax 170 pushl %eax 171 172 /* Enter paged protected Mode, activating Long Mode */ 173 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */ 174 movl %eax, %cr0 175 176 /* Jump from 32bit compatibility mode into 64bit mode. */ 177 lret 178ENDPROC(startup_32) 179 180no_longmode: 181 /* This isn't an x86-64 CPU so hang */ 1821: 183 hlt 184 jmp 1b 185 186#include "../../kernel/verify_cpu_64.S" 187 188 /* 189 * Be careful here startup_64 needs to be at a predictable 190 * address so I can export it in an ELF header. Bootloaders 191 * should look at the ELF header to find this address, as 192 * it may change in the future. 193 */ 194 .code64 195 .org 0x200 196ENTRY(startup_64) 197 /* 198 * We come here either from startup_32 or directly from a 199 * 64bit bootloader. If we come here from a bootloader we depend on 200 * an identity mapped page table being provied that maps our 201 * entire text+data+bss and hopefully all of memory. 202 */ 203 204 /* Setup data segments. */ 205 xorl %eax, %eax 206 movl %eax, %ds 207 movl %eax, %es 208 movl %eax, %ss 209 movl %eax, %fs 210 movl %eax, %gs 211 lldt %ax 212 movl $0x20, %eax 213 ltr %ax 214 215 /* 216 * Compute the decompressed kernel start address. It is where 217 * we were loaded at aligned to a 2M boundary. %rbp contains the 218 * decompressed kernel start address. 219 * 220 * If it is a relocatable kernel then decompress and run the kernel 221 * from load address aligned to 2MB addr, otherwise decompress and 222 * run the kernel from LOAD_PHYSICAL_ADDR 223 * 224 * We cannot rely on the calculation done in 32-bit mode, since we 225 * may have been invoked via the 64-bit entry point. 226 */ 227 228 /* Start with the delta to where the kernel will run at. */ 229#ifdef CONFIG_RELOCATABLE 230 leaq startup_32(%rip) /* - $startup_32 */, %rbp 231 movl BP_kernel_alignment(%rsi), %eax 232 decl %eax 233 addq %rax, %rbp 234 notq %rax 235 andq %rax, %rbp 236#else 237 movq $LOAD_PHYSICAL_ADDR, %rbp 238#endif 239 240 /* Target address to relocate to for decompression */ 241 leaq z_extract_offset(%rbp), %rbx 242 243 /* Set up the stack */ 244 leaq boot_stack_end(%rbx), %rsp 245 246 /* Zero EFLAGS */ 247 pushq $0 248 popfq 249 250/* 251 * Copy the compressed kernel to the end of our buffer 252 * where decompression in place becomes safe. 253 */ 254 pushq %rsi 255 leaq (_bss-8)(%rip), %rsi 256 leaq (_bss-8)(%rbx), %rdi 257 movq $_bss /* - $startup_32 */, %rcx 258 shrq $3, %rcx 259 std 260 rep movsq 261 cld 262 popq %rsi 263 264/* 265 * Jump to the relocated address. 266 */ 267 leaq relocated(%rbx), %rax 268 jmp *%rax 269 270 .text 271relocated: 272 273/* 274 * Clear BSS (stack is currently empty) 275 */ 276 xorl %eax, %eax 277 leaq _bss(%rip), %rdi 278 leaq _ebss(%rip), %rcx 279 subq %rdi, %rcx 280 shrq $3, %rcx 281 rep stosq 282 283/* 284 * Do the decompression, and jump to the new kernel.. 285 */ 286 pushq %rsi /* Save the real mode argument */ 287 movq %rsi, %rdi /* real mode address */ 288 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */ 289 leaq input_data(%rip), %rdx /* input_data */ 290 movl $z_input_len, %ecx /* input_len */ 291 movq %rbp, %r8 /* output target address */ 292 call decompress_kernel 293 popq %rsi 294 295/* 296 * Jump to the decompressed kernel. 297 */ 298 jmp *%rbp 299 300 .data 301gdt: 302 .word gdt_end - gdt 303 .long gdt 304 .word 0 305 .quad 0x0000000000000000 /* NULL descriptor */ 306 .quad 0x00af9a000000ffff /* __KERNEL_CS */ 307 .quad 0x00cf92000000ffff /* __KERNEL_DS */ 308 .quad 0x0080890000000000 /* TS descriptor */ 309 .quad 0x0000000000000000 /* TS continued */ 310gdt_end: 311 312/* 313 * Stack and heap for uncompression 314 */ 315 .bss 316 .balign 4 317boot_heap: 318 .fill BOOT_HEAP_SIZE, 1, 0 319boot_stack: 320 .fill BOOT_STACK_SIZE, 1, 0 321boot_stack_end: 322 323/* 324 * Space for page tables (not in .bss so not zeroed) 325 */ 326 .section ".pgtable","a",@nobits 327 .balign 4096 328pgtable: 329 .fill 6*4096, 1, 0 330